chore(profiles): sync GPU profiles with pcie_topology/NUMA from upstream (RUN-40173) (#212)

## What

Re-syncs the built-in GPU profiles via `hack/sync-profiles.sh` from
upstream NVIDIA/k8s-test-infra **`main`** (synced commit `497fa04`), and
improves the sync tooling.

The previous pin was the `v0.1.0` tag, which predates upstream's
**`pcie_topology`** block (PCI root complexes + per-device `numa_node`).
Without it, the mock backend's device-plugin can't report GPU→NUMA, so
`NodeResourceTopology` zones never get GPU resources and
NUMA/topology-aware scheduling can't be exercised on mock GPUs.
`pcie_topology` currently exists **only on upstream `main`** (no tagged
release yet).

## Changes

- **`hack/sync-profiles.sh`**:
- `DEFAULT_VERSION` now tracks **`main`** (the committed `builtin.yaml`
is the pinned artifact; each sync produces a reviewable PR diff).
Override with a tag/commit once upstream cuts a release.
- Accepts a **tag, branch, OR commit SHA**: blobless+sparse clone then
`git checkout <ref>` (the old `git clone --branch` only accepted
tags/branches, though the workflow advertises "tag or commit").
- The generated header records the **exact resolved commit** for
provenance: `# Source: NVIDIA/k8s-test-infra main (commit 497fa04)`.
- **`builtin.yaml`** (regenerated): all 7 profiles now carry
`pcie_topology` (a100/b200/gb200/**gb300**/h100/l40s/t4); `gb300` is
new.
- **`.github/workflows/sync-gpu-profiles.yml`**: fix version extraction
(`head -2` → `head -3`) — the `# Source:` line is line 3, so it
previously captured nothing.
- **`CHANGELOG.md`**: entry under `[Unreleased] → Changed`.

## Verification

- `hack/sync-profiles.sh` (default → `main`) ran clean → 7 profiles,
header pinned to `497fa04`.
- `builtin.yaml` parses as valid YAML (7 ConfigMap docs); each profile
contains `pcie_topology`.

## Context

Found while testing the mock backend on a real-NUMA EKS node: the real
NFD topology-updater produces a populated NRT (CPU zone `cap=8
alloc=7`), but GPUs don't appear in the zones because the rendered mock
profile had no `pcie_topology`. This sync is the prerequisite for
mock-backend GPU↔NUMA.

RUN-40173

Signed-off-by: Eliran Wolff <eliranw@nvidia.com>

Author: eliranw

.github/workflows/sync-gpu-profiles.yml

@@ -40,7 +40,7 @@ jobs:
         if: steps.diff.outputs.changed == 'true'
         id: version
         run: |
-          ver=$(head -2 deploy/fake-gpu-operator/templates/profiles/builtin.yaml | grep '# Source:' | sed 's/.*k8s-test-infra //')
+          ver=$(head -3 deploy/fake-gpu-operator/templates/profiles/builtin.yaml | grep '# Source:' | sed 's/.*k8s-test-infra //')
           echo "version=$ver" >> "$GITHUB_OUTPUT"
 
       - name: Create PR

CHANGELOG.md

@@ -12,9 +12,18 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ### Changed
 
+- Built-in GPU profiles re-synced from NVIDIA/k8s-test-infra `main` (commit
+  `497fa04`): each profile now includes a `pcie_topology` block (PCI root
+  complexes with per-device `numa_node`), and a `gb300` profile is added. This
+  is what lets the mock backend report per-GPU NUMA affinity. (RUN-40173)
+- `hack/sync-profiles.sh`: default source bumped `v0.1.0` → `main`; now resolves
+  a tag, branch, or commit SHA (was tags/branches only) and records the resolved
+  commit in the generated `# Source:` header. (RUN-40173)
+
 ### Fixed
 
 - `device-plugin` injects `NODE_NAME` so non-DRA pods can run the fake `nvidia-smi`. ([#191](https://github.com/run-ai/fake-gpu-operator/issues/191))
+- `sync-gpu-profiles` workflow read the synced version with `head -2`, but the `# Source:` line is line 3, so the PR title/commit version was always empty. (RUN-40173)
 - CI `e2e-upgrade (latest-main)` lane no longer deadlocks resolving its baseline
   chart. It now walks `--first-parent` main commits (only those publish a
   `0.0.0-<sha>` chart, so merges no longer fill the window with unpublished